Temporal variation in rates, such as growth, survival and fecundity, is influenced by life history tradeoffs (which tend to cause negative correlations through time) and correlated environmental variation (which tends to lead to positive correlations through time). Relative to uncorrelated "white noise", positive correlations tend to exaggerate the effects of environmental stochasticity, whereas negative correlations tend to dampen its effects. Depending on the timing of resource allocation relative to data collection, effects of life history tradeoffs and correlated environmental variation may be detected as correlations of vital rates in the same year, or serial correlations between rates in successive years. I developed a novel extension of generalized linear mixed models (GLMMs) to statistically estimate correlations and serial correlations among vital rates through time. I use this method to quantify the extent to which demographic time series for perennnial plants are dominated by negative vs. positive correlations. I also evaluate how modeling the variance sturcture simulataneously with parameter estimation affects estimates of annual vital rates and their variances through time.
Results/Conclusions
Analysis of a 25-year demographic time series for Astragalus scaphoides, an iteroparous perennial plant that flowers in alternate years, revealed weak positive correlations among survival and fecundity, but strong negative serial correlations. Accounting for this correlation structure changed estimates of annual vital rates, relative to GLMMs without correlated variances, or simple estimates of vital rates from frequences of plant fates. Simulation studies indicated that parameter estimates (including serial correlations) from shorter time series were unbiased, albeit imprecise. I am in the process of applying this approach to long-term time series for other perennial plants, with differing levels of evidence for demographic costs of reproduction: Cypripedium acaule (Primack and Stacey 1998), Silene spaldingii (Lesica and Crone 2007), and Sarracenia purpurea (Gotelli and Ellison 2002).
In general, demographic studies for both plants and animals have tended to assume uncorrelated environmental variation, or, more recently, to include correlations among rates within years but not serial correlations between years. The examples in this presentation indicate that it is both important and feasible to account for the full correlation structure, at least when modest (10-20 year) demographic time series are available. Further, because estimates from shorter time series appear to be unbiased, the hierarchical GLMM framework allows for the possibility of combining multiple shorter time series to estimate patterns of demographic variation across life history guilds, or similar life history groups.